基于卷积神经网络的监控场景下车辆颜色识别  

Vehicle Color Recognition Using Convolutional Neural Network for Urban Surveillance Images

作　　者：张强[1] 李嘉锋[1] 卓力[1] ZHANG Qiang LI Jia-feng ZHUO Li(Signal ＆ Information Processing Laboratory, Beijing University of Technology, Beijing 100124, China)

机构地区：[1]北京工业大学信号与信息处理研究室,北京100124

出　　处：《测控技术》2017年第10期11-14,共4页Measurement & Control Technology

基　　金：国家自然科学基金资助项目(61531006;61372149;61370189;61471013);北京市属高等学校高层次人才引进与培养计划项目(CIT&TCD20150311;CIT&TCD201404043);北京市自然科学基金资助项目(4142009;4163071);北京市教育委员会科技发展计划资助项目(KM201410005002;KM201510005004);北京市属高等学校人才强教计划资助项目PHR(IHLB)

摘　　要：颜色是车辆识别中广泛应用的主要线索之一,在智能交通系统中扮演着重要的角色。受光照变化、噪声、环境等复杂因素的影响,传统的车辆颜色识别方法难以取得理想的识别效果。利用卷积神经网络(CNN)的优越识别性能,提出了一种基于卷积神经网络的监控场景下车辆颜色识别方法。基于传统的CNN原理设计了车色识别专用深度网络架构,直接通过CNN学习基于颜色分布的分类模型。与其他基于深度学习的车色识别方法相比,提出的用于车色识别的专用深度网络,具有参数少、识别速度快、识别精度高等优点。实验结果表明,在Chen等公布的标准数据集上,与最新的研究成果相比,平均识别精度提高约0.77%,识别速度提高14倍左右。Color is an important visual cue for vehicle color recognition, which is one of essential parts in intel- ligent traffic system （ ITS）. The accuracy of the conventional vehicle color recognition in complex environments cannot achieve a satisfied result effect by the influences of various complex environmental factors such as illu- mination, weather, noise and etc. A high-accuracy vehicle color recognition method using convolutional neural network （CNN） for urban surveillance images is proposed. Based on the traditional CNN principle, the special depth network architecture for vehicle color recognition is designed, and the classification model based on color distribution is learned directly through CNN. Compared with the other deep-learning-based methods, the pro- posed method has the advantages of fewer parameters, higher speed and higher precision of recognition. The ex- perimental results on the public database demonstrate that the recognition accuracy of the proposed method can achieve superior performance over the state-of-the-arts methods. The average recognition accuracy is improved by about 0.77%, and the recognition speed is increased by about 14 times.

关 键 词：车辆颜色识别 卷积神经网络 图像处理 智能交通系统 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]